March 2024 iteration of the DeepSeek V3 685B MoE model.
NVIDIA's open multimodal mixture-of-experts model, with 12B active parameters out of 120B total.
No shared demos for these models yet.
No captured outputs for these models yet.
Both DeepSeek V3 (0324) and Nemotron 3.5 Super VL are capable AI models. The best choice depends on your specific use case: consider pricing, context window, speed, and which capabilities (vision, reasoning, audio) you need.
Nemotron 3.5 Super VL has a lower blended cost. DeepSeek V3 (0324): $0.29 input / $1.14 output. Nemotron 3.5 Super VL: $0.10 input / $0.40 output.
Nemotron 3.5 Super VL has a larger context window: DeepSeek V3 (0324) supports 128K tokens vs Nemotron 3.5 Super VL at 262K tokens.
Nemotron 3.5 Super VL supports vision/image input, but DeepSeek V3 (0324) does not.
Key differences: Nemotron 3.5 Super VL is significantly cheaper; only Nemotron 3.5 Super VL supports vision input; only Nemotron 3.5 Super VL offers extended reasoning mode. Compare full specs on this page.
If cost is your priority, choose the cheaper option. If you need the highest intelligence for complex tasks, pick the higher-scoring model. For long documents or codebases, choose the larger context window. You can try both DeepSeek V3 (0324) and Nemotron 3.5 Super VL for free on idapt.app to see which performs better for your specific needs.